Breast Density Analysis Based On Glandular Tissue Segmentation And Mixed Feature Extraction

In recent years,breast cancer has become one of the major diseases threatening women's health.Relevant studies have shown that breast density is closely related to the risk of breast cancer,and is listed as one of the indicators of breast cancer risk.Therefore,the classification of breast density is of great significance.This paper proposes a breast density analysis method based on gland tissue segmentation and mixed feature extraction,and designs a breast density classification system based on this method,which includes the following four modules:preprocessing,glandular tissue segmentation,feature extraction and density classification.The specific research contents of this paper are as follows:1.In the preprocessing module,we propose a mask-based segmentation method,which can set different geometries as a mask according to the needs,and use the mask shielding effect to cover labels and pectoral muscles in the breast image,so as to remove these interfering information.Then,the grayscale stretching function was used to stretch the breast images of different types of glandular types,so that the contrast of glandular tissue was enhanced and the brightness of adipose tissue was suppressed,presenting the most favorable state for segmentation.2.In the glandualr tissue segmentation module,we propose an automatic threshold segmentation method based on BP neural network.To begin with,we obtain the segmentation thresholds of breast image by means of modulation and take these thresholds as the expected output of BP neural network.Then,the statistical features of the breast image were calculated and the texture features of the image were extracted using the GLCM as the input of the BP neural network.Finally,the BP network is trained to automatically predict the segmentation threshold of the image and realize the automatic segmentation of glandular tissue.3.In the feature extraction module,we extract three types of feature vectors and mix them.Firstly,GLCM was used to extract the texture features of glandular tissue,and then three statistical features(mean,skewness and kurtosis)were calculated.Secondly,the area method is used to calculate the estimated breast density and add it as one of the features to the existing feature vectors to form a mixed feature vector.Finally,the mixed feature vectors extracted from different databases are used to train the classifier parameters,so that the classification accuracy of different database samples can be obtained according to the predicted output of the classifier.4.In the density classification module,we use SVM and ELM to perform the breast density classification experiment.In this paper,the classification of breast density was compared among three databases,namely,the MIAS,the mixed Database(Gansu Cancer Hospital Database and MIAS),and the DDSM.Based on SVM and ELM,by setting the density of three types of network output,with 96.19%and 84.17%respectively in the MIAS forecast classification accuracy,in the hybrid database respectively 95.01%and 87.44%of the predicted classification accuracy,and samples of DDSM database can be divided into four types of density is 96.35%and 77.10%of the predicted classification accuracy,the experimental results show that SVM is more accurate in the density classification system in this paper,and using it as the main classifier makes our density classification system have higher accuracy and better generalization performance.